[Feature] timelimit faces for face recognition

[derestle-htwg]

I have searched the existing feature requests, both open and closed, to make sure this is not a duplicate request.

• Yes

The feature

My younger kids looks like her older sibling some years ago.
That leads to the situation that the younger kid always gets tagged as the older sibling.

Make it possible that automatically tagging a person can be time limited.
E.g. I mark my two kids with a face tagging time limit of 1 year, meaning that after one year they are not tagged anymore.

Instead i have to add a new reference picture and then they get tagged from that reference for another year.

Platform

• Server
• Web
• Mobile

[arthurnl]

This request closely matches a problem I encounter while organising a historical family photo archive spanning several decades. I would like to build on the idea of adding a new reference each year: retain the confirmed references as a year-by-year collection for each person, and preferentially use references from the same or nearby years when matching a face.
Would the following be a useful direction to develop within this discussion?
Year-by-year reference faces for recognising people across decades
The same person may appear in my archive from infancy through adulthood, with substantial changes in facial appearance. The photographs have an established year and chronological order. Exact dates within a year are not required; establishing that chronology is outside this proposal’s scope.
Proposed feature
Maintain human-confirmed reference faces for each person, organised by year, and use those references to support recognition as that person’s appearance changes over time.
As successive years of the archive are processed, the user confirms identities and adds reference faces for those years. This progressively builds a chronological reference collection for each person.
When recognising a face, Immich could preferentially compare it with confirmed references from the same or nearby years, while retaining access to the person’s wider reference collection.
Example workflow

1. Process a year of photographs.
2. Review detected faces and confirm their identities.
3. Retain selected confirmed faces as that person’s references for that year.
4. Use the growing reference collection when processing subsequent years.
   For example, confirmed references of a child from 1990 and 1991 could help identify that child in photographs from 1992. References added in later years would progressively document the person’s changing appearance.
   Relationship to existing recognition
   Immich already compares multiple face embeddings. The proposed addition is an explicit, human-confirmed reference collection organised by person and year, with year-aware reference selection or weighting.
   The proposal builds on existing embeddings and does not require a separate facial age-estimation model. Birth-date exclusion is existing functionality and is not the focus of this request.
   Questions to investigate

• How should users select and maintain confirmed yearly references?
• How should confirmed references be distinguished from automatic assignments?
• How should temporal proximity influence matching, particularly when some years have no references?
• Does this improve recognition compared with Immich’s current matching approach?
  Improved accuracy is a hypothesis to validate against a manually checked historical archive.
  Scope and contribution
  The feature should support incremental enrichment as more years are processed. Its development must remain independent of ongoing archive ingestion.
  I would be interested in helping validate the approach and, if technically justified, contributing to a proof of concept or upstream implementation.